Baird's $1,250 AMD Target Rests on Two Numbers That Only Multiply Against a Smaller Market

by Nikolai Petrov · August 15, 2026

Technology Reporter, Frontier Tech Wire

Advanced Micro Devices (Nasdaq: AMD) closed Friday, August 14 at $514.39, up 6.50%, according to both stockanalysis.com and MarketBeat, which agree to the cent. Two other providers do not: StockStory puts the close at $514.13, also at +6.5%, while TipRanks reports $512.93 and +6.19%. The direction and the rough magnitude are not in dispute; the last decimal is. Against Thursday's close of $483.01, a move of about 6.5% is the arithmetic that reconciles, and the widely circulated figure of $507.77 and +5.13% that appeared in Friday coverage was an intraday print taken before the bell, not a close. StockStory, which reports both, has AMD up about 5.1% in the afternoon session before finishing at +6.5% — consistent with $507.77 having been a level the stock passed through rather than one it ended at.

The reason given for the move, in a great deal of Friday's coverage, was a research note from Robert W. Baird carrying a Street-high $1,250 price target on AMD. That target is real. The date attached to it in Friday's coverage was not. Baird analyst Tristan Gerra doubled his target to $1,250 from $625 while maintaining an Outperform rating on July 24, 2026, according to price-target.com and Investing.com, both of which timestamp the note to that day. Friday was three weeks later.

The confusion is traceable. 24/7 Wall St. published a piece at 8:29 a.m. ET on Friday describing Gerra as having doubled his target "to $1,250 from $625 while maintaining an Outperform rating" and setting "the Street-high target across all sell-side analysts covering the stock." All of that is accurate. What the piece does not do is say when the note was published, and its reference price of $483.01 was Thursday's close, because the article ran ninety minutes before the opening bell. A later 24/7 Wall St. piece the same afternoon carried the headline framing of AMD climbing on "Baird's $1,250 call." A three-week-old target restated in a Friday-morning article is not a Friday catalyst, and we could find no source showing Baird taking any action on AMD on August 14.

The prior target checks out, and this was a raise rather than a resumption of coverage. Stockanalysis.com's analyst page for Gerra records him maintaining a Buy on AMD on May 6, 2026 with the target raised to $625 from $300. The July 24 note took $625 to $1,250. The same page identifies Gerra as a Robert W. Baird analyst covering semiconductors, electronic components, computer hardware and software infrastructure, with 934 ratings logged. Name, firm, prior target and the raise-not-resumption question all hold.

What did move AMD on Friday is more mundane and better documented. The company completed its largest dollar bond offering, $4.75 billion across four tranches: $1.25 billion at 4.600% due 2029, $1.5 billion at 5.000% due 2031, $1 billion at 5.250% due 2033 and $1 billion at 5.500% due 2036, per Tom's Hardware and GuruFocus, which describe the same structure. Tom's Hardware dates the announcement to Thursday; TipRanks describes the sale as closing Friday with demand strong enough to price the longer-dated tranches below initial guidance. AMD's preliminary 424B5 prospectus supplement, filed August 13, states the use of proceeds as "general corporate purposes, which may include the repayment of debt," with the principal amounts and coupons still blank at the preliminary stage. Tom's Hardware notes AMD carried roughly $3.2 billion of total debt at the end of the second quarter, $875 million of it current, against $13.1 billion of cash and short-term investments. StockStory attributed Friday's gain principally to the bond pricing. Separately, Bank of America's Vivek Arya published a note on Thursday, August 13, per Benzinga, lifting his 2030 server CPU market estimate to $210.6 billion from about $170 billion previously and modelling the market growing to that level from $61.4 billion in 2026. He called AMD "our top CPU pick on its wide breadth of portfolio and dual leadership – highest-frequency (compute/head node rack) and highest core/thread count (agentic AI)."

With the provenance settled, the Baird model itself is worth taking apart, because it rests on exactly two load-bearing numbers and they do not obviously multiply together. Investing.com's July 24 write-up states that "Baird now models $147 billion in AI GPU revenue by 2030, assuming a 15% market share." 24/7 Wall St., on July 30, rendered the same claim as data center AI GPU revenue of roughly $147 billion by 2030 "as AMD captures about 15% market share in enterprise and hyperscale AI accelerators." Its August 14 piece words it a third way: "AI GPU platform revenue reaching $147 billion by 2030, assuming roughly 15% share of the data center AI accelerator TAM." That is the same outlet, two weeks apart, naming two different denominators — one a subset of the accelerator market, the other the whole of it. A revenue figure and a share figure imply a market size, and that implied market size is checkable.

It is checkable because AMD itself published a 2030 market forecast the day before the Baird note, and the number came from the chief executive on stage rather than from a broker's summary of her. At Advancing AI 2026 on July 23, Lisa Su said AMD "now expect[s] that by 2030, the AI accelerator market is going to reach about $1.4 trillion," up from the roughly $500 billion the company had forecast a year earlier, according to Investing.com's transcript of the keynote. The Next Platform, covering the same session, reports AMD putting training and inference together on "more than a 45 percent compound annual growth rate between 2025 and 2030 to hit more than $1.4 trillion dollars." Su put the server CPU market at "over $200 billion by 2030" — a figure Bank of America's summary of the event rendered as $220 billion, against a prior $120 billion. AMD's own press release for the event states only the wider number: AI is "driving AMD's TAM to ~$2 trillion in 2030," for total compute. Baird, per Investing.com, describes that $2 trillion as growing from $365 billion in 2025, a roughly 40% compound annual growth rate.

Run the arithmetic against that $1.4 trillion accelerator figure, bearing in mind throughout that the $147 billion and the 15% reach us only as secondary paraphrase and not as quotations from the note. AMD earning $147 billion of a $1.4 trillion market is a 10.5% share, not 15%. Fifteen percent of $1.4 trillion is $210 billion, some $63 billion above the revenue number in circulation. Working the other direction, $147 billion at exactly 15% implies an accelerator market of $980 billion in 2030 — about 30% smaller than the figure Su gave a day earlier. Which way that cuts depends on which paraphrase one takes. On 24/7 Wall St.'s Friday wording, where the 15% is share of "the data center AI accelerator TAM," the two denominators are the same object and the inconsistency is real. On the same outlet's earlier wording, where the 15% is share of enterprise and hyperscale accelerators only, the denominator is narrower than AMD's and the gap may close entirely. The figures as relayed are internally consistent only against a denominator materially below AMD's own — but we cannot establish from any fetched source that Baird was using AMD's denominator.

There are several innocent explanations, and we cannot choose between them because we have not seen the note. Baird may be using its own accelerator TAM rather than AMD's. The 15% may refer to unit or wafer share rather than revenue share, or to a defined subset such as merchant GPUs sold to enterprise and hyperscale buyers, excluding the custom silicon that hyperscalers design in-house — which would be a smaller denominator and could reconcile the two. The wording of the revenue figure also differs between outlets: Investing.com says "AI GPU revenue," while 24/7 Wall St. says "AI GPU platform revenue," and a rack-scale platform number that swept in CPUs and networking would be a larger figure than a GPU-only number. Three renderings of one sentence in a note none of us has read is a thin basis for a verdict, and we are not returning one.

Set the target aside and look at what AMD is shipping. In the second quarter of 2026, reported August 4, AMD posted record revenue of $11.5 billion, up 50% year over year, with data center revenue of $6.7 billion — more than double the prior year, 58% of the company, and $2.1 billion of operating income against a $155 million loss a year earlier, per Investing.com's read of the slides. GAAP gross margin was 54% and non-GAAP 56%; GAAP diluted EPS was $1.38 and non-GAAP $1.66. Third-quarter guidance is approximately $13 billion plus or minus $300 million, about 41% year-over-year growth, with non-GAAP gross margin around 56%. For all of 2025, AMD reported revenue of $34.6 billion and data center segment revenue of $16.6 billion. Those are the bases from which $147 billion has to be reached. AMD's entire data center segment in 2025, EPYC server CPUs included, was $16.6 billion; growing that to $147 billion by 2030 is 8.9 times over five years, a compound rate of about 55% a year, and the comparison flatters the model because $147 billion is a GPU-only figure while $16.6 billion is not. Against the annualized third-quarter guide of roughly $52 billion, $147 billion is 2.8 times the whole company. Against 2025 revenue of $34.6 billion, it is 4.2 times.

One cross-check comes closer to reconciling. At its Financial Analyst Day on November 11, 2025, AMD set a three-to-five-year target of a data center AI revenue compound annual growth rate of more than 80%, alongside company revenue growth above 35%, data center growth above 60%, non-GAAP operating margin above 35% and non-GAAP EPS above $20. If AMD's AI GPU revenue compounds at exactly 80% for the four years from 2026 to 2030, it multiplies by about 10.5 — which means $147 billion in 2030 implies roughly $14 billion of AI GPU revenue in 2026. AMD does not break Instinct out from EPYC inside the data center segment, and neither the release nor the call gives that split, so we cannot confirm the 2026 figure. But $14 billion inside a segment running at roughly $27 billion annualized is at least an arithmetically coherent starting point rather than an absurd one.

Share, though, is a claim about other people's revenue, and this is where a 2030 forecast has to survive contact with current run rates. Nvidia (Nasdaq: NVDA) reported data center revenue of $75.2 billion in its fiscal first quarter of 2027 alone, up 92% year over year, on total revenue of $81.6 billion, with second-quarter guidance of $91.0 billion plus or minus 2%. That is roughly $300 billion a year of data center revenue at the most recent quarter's rate, from a company whose market capitalization stockanalysis.com puts at $5.45 trillion against a Friday close of $225.16. AMD's entire company is running at about a sixth of Nvidia's data center segment.

The custom-silicon side is the other claimant. Broadcom (Nasdaq: AVGO) reported AI semiconductor revenue of $10.8 billion in its fiscal second quarter ended May 3, up 143% year over year, and guided the third quarter to $16.0 billion, growth of more than 200%. Chief executive Hock Tan attributed it to "increasing demand for custom AI accelerators and AI networking." At the guided rate that is a $64 billion annual run rate for hyperscaler ASICs and the networking sold with them. Add Nvidia's data center run rate and Broadcom's AI run rate together and the two account for roughly $365 billion a year as of mid-2026. Two caveats, and they pull in opposite directions: both figures bundle networking alongside accelerators, which inflates them against a pure accelerator TAM, and the two companies are not the whole market, which deflates them. The figure is a rough marker, not a measurement. It is also a coincidence that Baird's 2025 total-compute starting point is likewise $365 billion; the two numbers measure different things and should not be read as confirming one another.

Getting from that roughly $365 billion to $1.4 trillion by 2030 requires about 3.8 times growth in four and a half years, near 35% compounded annually. AMD's own framing is more demanding still: run Su's more-than-45% compound rate back from $1.4 trillion and it implies a 2025 accelerator market of roughly $220 billion or less — our arithmetic on her stated growth rate and endpoint, not a figure AMD published. Either way, that is the assumption underneath the assumption. AMD taking 15%, or 10.5%, of a $1.4 trillion market does not require it to take a single dollar away from Nvidia; it requires the market to roughly quadruple and AMD to capture a modest slice of the increment. Whether that is more or less demanding than a share-shift story depends entirely on what one believes about the size of the pie, and nothing published this week settles it.

Baird's own coverage suggests the model is TAM-driven rather than share-shift-driven, which is a point in its favour on internal consistency. On May 21, 2026, Gerra raised his Nvidia target to $500 from $300 with an Outperform rating, following Nvidia's fiscal first-quarter results; TipRanks described it at the time as a Street high. The large market forecast that accompanied it is not Baird's own. Investing.com's account of the note attributes it to the company: "Nvidia expects AI infrastructure spending to go from over $1 trillion in 2027 to $3 trillion to $4 trillion annually by 2030 as agentic AI proliferates across all industries." That is a forecast Gerra was relaying, not one we can show he authored, and we have described it wrongly if we imply otherwise. Stockanalysis.com's analyst page also records a $630 target on Broadcom dated June 3; we found no independent write-up of that note and cannot say where it sat against the Street. A firm carrying targets that high on the incumbent, the challenger and the leading custom-silicon vendor at once is not forecasting a share war. It is forecasting a very large market. The $1.4 trillion accelerator figure would sit inside a $3 trillion to $4 trillion infrastructure figure at a plausible silicon-content ratio — but since the larger number is Nvidia's rather than Baird's, that is a consistency check on the industry's narrative, not on Baird's model.

Product cadence is the part of the thesis with the most public detail. At Advancing AI on July 23, AMD said its Helios rack-scale systems had entered production, launched the Instinct MI400 series and sixth-generation EPYC "Venice" CPUs, and laid out MI500 for 2027 and MI600 for 2028. On the August 4 earnings call, Lisa Su said Helios "is now in production with initial shipments on track to begin later this quarter and ramp through the fourth quarter and into 2027," that customer pull was "tracking ahead of our initial forecast," and that AMD plans "to launch a new rack scale AI platform every year with each generation delivering significant performance, efficiency and TCO gains." She also said AMD now expects data center segment revenue to more than double year over year in 2027. An annual rack cadence sustained for four consecutive generations is the operational precondition for any 2030 share number, and AMD has executed one such transition so far. Customer commitments are named but mostly forward-dated. AMD's Advancing AI release says Anthropic will deploy up to 2 gigawatts of Instinct MI455X GPUs in Helios systems, that OpenAI expects to bring Helios online beginning in the fourth quarter of 2026 with deployments accelerating through 2027, and that Meta is validating sixth-generation EPYC platforms and testing workloads on Helios racks, with Microsoft and Oracle also named. AMD's own July 22 release on that partnership says deployment of the first gigawatt begins in the first half of 2027 — timing Su repeated on the August 4 call — and that AMD "has committed to make a strategic equity investment of up to $5 billion in Anthropic." That is a ceiling on a commitment, not money already deployed. Announced gigawatts are not revenue until they are installed and paid for, and almost none of this capacity is in the ground today.

Supply is the constraint that a 2030 model cannot assume away, and management addressed it directly. On the August 4 call Su said: "I would say the server CPU supply chain is tight right now, and it has been tight for the first half of the year because much of this demand was unforecasted." She added that "as we get into 2027, the demand is better forecasted. And so we would expect that the 2027 server supply situation should be better than '26." On memory she said AMD is "working very, very closely with our memory partners across the board on both our GPU, HBM memory as well as just the general memory for the systems across our data center business," and that "there is very good visibility into HBM allocation for what we expect to deliver in 2027." On wafers she noted that because AMD uses chiplet technology, "we actually ramp in fewer wafers in the new node." Visibility into 2027 allocation is a genuine disclosure; it is also four years short of 2030, and we were unable to source a primary supplier statement covering HBM or advanced packaging capacity out to the end of the decade.

The dispersion around the target is wide, and the aggregators do not agree on where the bottom is. Stockanalysis.com, priced off Friday's close, shows 52 analysts with an average target of $612.84, a high of $1,250 and a low of $365. MarketBeat, same close, shows 45 analysts, an average of $546.87, the same $1,250 high and a low of $235. We could not date the $235 to any note we were able to fetch, and MarketBeat's list of recent rating actions included a Barclays Underweight that, when we followed it, turned out to be from July 2017 — so treat the bottom of that range as unverified. The lowest target we could date and confirm is HSBC's, which downgraded AMD to Hold from Buy on May 4, 2026 while raising its target to $340 from $335, citing a re-rating from about 19 times 2027 earnings to about 33 times and "limited room for earnings upside due to capacity constraints." That $340 sits below the $365 stockanalysis.com prints as its own low, which suggests the aggregator has stopped counting HSBC in its range — a reason to treat both published lows with caution rather than either as the floor. In the middle, after Advancing AI, Jefferies went to $640 from $515 and Bank of America to $620 from $560, both Buy, per Investing.com's July 24 round-up; Bernstein's Stacy Rasgon raised to $650 from $525 on August 13 and Argus's Jim Kelleher to $625 from $450 on August 6, per stockanalysis.com. Above them, and on the same day as Baird, UBS's Timothy Arcuri went to $730 from $700, per TheStreet; KeyBanc had taken its target to $725 from $530 on July 17, briefly a Street high in its own right, per Yahoo Finance; and MarketBeat records Rosenblatt at $700 from $665 and Wells Fargo at $700 from $615. Baird's $1,250 is still not the top of a cluster — it is roughly 1.7 times the next-highest target we could confirm — but it is not the lonely outlier that a bare "Street-high" framing implies, and several firms reached the low $700s without needing a $147 billion line item.

One piece of scale arithmetic, since the target is a per-share number. AMD has about 1.63 billion shares outstanding and a market capitalization of $839.73 billion at Friday's close, per stockanalysis.com, which also puts the trailing P/E at 131.29 and the forward P/E at 46.32. At $1,250 the same share count implies a market capitalization of roughly $2.04 trillion — against Nvidia's $5.45 trillion today. That is our arithmetic on a reported target, not an endorsement of it, and it says nothing about whether $1,250 is right.

The week's other AI-financing verdict pointed the opposite way in the same session. Broadcom fell 5.94% on Friday to close at $392.99, per stockanalysis.com, on a Bank of America credit note flagging roughly $370 billion tied to the XPV chip-financing platform it built with Apollo and Blackstone. That note, like Baird's, was not a Friday event either: BofA cut Broadcom's issuer and bond ratings to Marketweight on Tuesday, August 11, and Friday's coverage recirculated it. As this desk reported yesterday, the $370 billion figure is described by different outlets — and by the same outlet on different days — as either a maximum residual value guarantee ceiling or a senior debt balance, and those are not the same claim. We did not resolve it then and we are not resolving it here. The point of the juxtaposition is narrower: on one Friday the market marked up a company on a 2030 demand forecast and marked down another on a 2029 financing structure, and neither the forecast nor the structure is a document the public can read.

What is knowable today is a short list. AMD's 2026 revenue run rate, its segment margins, its guided third quarter, its published roadmap and the gigawatt commitments it has announced are all on the record. AMD's own 2030 TAM of $1.4 trillion for accelerators is on the record as a company forecast, in the chief executive's own words on stage. What is not knowable is the denominator Baird is actually using, whether the 15% is revenue or unit share, whether $147 billion is GPU-only or platform revenue, what AMD's Instinct revenue is in isolation, and whether HBM and advanced packaging supply exist in 2030 at the volumes any of these forecasts assume. The next hard data points are dated: Nvidia reports fiscal second-quarter results on Wednesday, August 26 at 5:00 p.m. ET for the quarter ended July 26, and Broadcom on Wednesday, September 2. Those filings will move the denominator by more than any 2030 model will.

This article is news and analysis, not investment advice. Price targets and ratings are reported as attributed facts and are not adopted or endorsed here. Figures are as reported by the sources listed; where providers conflict, we have said so rather than picked one.